TECH FOCUS — CMP Monitoring slurry stability to reduce process variability John P. Bare and Travis A. Lemke, FSI International As chemical mechanical planarization (CMP) becomes more widely used for polishing during microcircuit manufacture, increasingly sophisticated process control procedures are being developed and implemented. One of the key areas is measurement of slurry quality. Regardless of the slurry type or source, maintaining consistent slurry quality or "health" is crucial to maximizing device yields.1 A slurry's health refers to its stability and contamination level. A healthy slurry, therefore, is one that maintains its polishing performance while resisting changes in settling and agglomeration tendencies and is free of contamination, including slurry particle aggregates and agglomerates, foreign particles, absorbed carbon dioxide, and ionic contaminants. Dynamic measurements of slurry properties, coupled with defined control limits, can be used to monitor and control health. The multipart slurries used by the semiconductor industry often require mixing or dilution before use. Oxide polishing slurries are commonly purchased in concentrated form and diluted with water on-site to minimize shipping and labor costs, while some multipart tungsten polishing slurries must be combined shortly before use because of their short postmix lifetime. In addition, settling characteristics of the abrasive may change after it is combined with the oxidizer. Occasionally slurry users or manufacturers measure some physical property unique to a specific slurry, but normally pH, specific gravity, percent solids by weight, and slurry particle size distribution are measured after mixing for both oxide and tungsten slurries as part of CMP process control. Some of these properties are more relevant than others as a monitor of slurry health, and other parameters not commonly monitored may be more useful. After presenting an overview of the relevant instrumentation, this article describes a study that evaluated the use of density, pH, zeta potential, and particle-size distribution as indicators of slurry health. Slurry Analysis Instrumentation In addition to choosing which properties to monitor, slurry users must decide whether to use on-line or off-line measurement instruments. On-line monitoring sounds attractive because it provides real-time data. Unfortunately, many analytical instruments are not easily adapted to on-line sampling, requiring rigorous maintenance and calibration for proper operation in such applications. The analytical instruments used for slurry quality control range from simple and inexpensive to complex and costly. Table I lists some of these instruments along with their capabilities. Measurement Instrument Analysis Location Density/specific Densitometer On-line gravity or off- line Percent solids Balance and Off-line oven Component Titrator Off-line assay pH pH probe and On-line meter or off- line Viscosity Viscometer Off-line Ionic ICP-MS and Off-line contamination GFAA Zeta potential Zeta potential Off-line analyzer Particle-size Particle-size Off-line distribution analyzer Table I: Slurry analysis instruments and their capabilities. As the table indicates, densitometers are used to measure density or specific gravity. Off-line measurement is easy to perform, while on-line measurement requires periodic instrument maintenance to ensure reliable results. Percent solids is normally analyzed off-line. For oxide slurries, the principal component of which is silica, this is a straightforward measurement following evaporation to dryness. However, because they contain dissolved solids with low decomposition temperatures, many tungsten slurries require a high-temperature drying procedure that completely decomposes these components in order to achieve repeatable results.2 Component assays monitor the concentration of dissolved solids (potassium iodate, ferric nitrate, and so forth) in the chemical component of slurries. The most common technique is off- line titration for some specific ingredient. Unfortunately, many commercial slurries contain proprietary ingredients whose composition and concentration are not made available to users, making measurement, and thus monitoring, more challenging. Another commonly measured slurry property is pH, although such data may not provide meaningful information for slurries that are chemically buffered. Even in unbuffered oxide slurries, silica can act as a buffering agent. However, pH measurements can indicate contamination levels in slurry. Oxide slurries in contact with air absorb carbon dioxide, and the resulting decrease in pH can affect the CMP process. The precision and reliability of a pH meter can be affected by several factors. The potential unreliability of standard pH probes for on-line monitoring is due in part to clogging of the porous probe membrane by small abrasive particles and in part to the amount of calibration and maintenance required. Since pH is dependent on temperature, pH readings also must be temperature compensated. Significant changes in pH can reduce the absolute charge on slurry particles, leading to particle agglomeration. Consequently, the relationship between zeta potential and pH can provide useful information for monitoring slurry health. Other properties, such as viscosity and ionic contamination, appear on most slurry suppliers' product information sheets. They are occasionally measured by slurry users as well but rarely are used as routine monitors of slurry health. Particle-size distribution is the most recent focus area for monitoring slurry health. Unfortunately, different particle-size analyzers generate different distributions for the same slurry. There are three major factors that contribute to these differences: (1) the various commercially available instruments use different physical principles to measure particle size; (2) slurry particles are irregularly shaped, in contrast to the polystyrene spheres used to calibrate many of the instruments; and (3) each instrument uses a different algorithm or statistical treatment of the raw data. Such differences make it difficult to compare data from the various analyzers available to the CMP community. Furthermore, there is no consensus on whether average particle size is more or less important than the number of large agglomerated particles. Generally, larger particles increase removal rates during CMP processing, but very large agglomerates may contribute to wafer scratching.3,4 Experimental Procedures Four commercially available slurries (listed in Table II) were used in the evaluation of density, pH, zeta potential, and particle-size distribution as indicators of slurry health. The four products represent the main categories of oxide and tungsten slurries used in volume wafer production. All are two-part slurries, which provided an opportunity to measure the ability of the various analytical techniques to detect mixing error. In order to test this capability, "high-solids" and "low-solids" slurries were prepared such that the percentage of abrasive was 20% higher or 20% lower than the recommended ratio. (Since some slurries contain dissolved solids in the oxidizer component, their total percent solids may vary.) Slurry Manufacturer Active Abrasive Type/Trade Chemical Name Oxide polishing slurries Cab-O-Sperse Cabot Potassium Silica SC-1 hydroxide Klebosol Solution Ammonium Silica 30N50 Technology hydroxide Tungsten polishing slurries Semi-Sperse Cabot Ferric nitrate Alumina W-A400/FE- 400 Granite 14 Rodel Potassium Alumina iodate Table II: Slurries used in the evaluation of density, pH, zeta potential, and particle-size distribution as indicators of slurry health. In addition to the high- and low-solids samples, a sample of each slurry was prepared at the volumetric mix ratio recommended by the manufacturer (Table III). The diluent (water) or oxidizer was added slowly. Physical properties were measured within 30 hours of sample preparation. Density was determined using a volumetric flask: A tared 100-ml flask was filled with slurry and weighed to the nearest 0.001 g. The pH levels were measured on samples that had been equilibrated in a constant-temperature bath at 20°C using a meter that was calibrated by the standard two-point technique before each session and checked for drift after each session. Percent nonvolatile solids was measured following evaporation to dryness. For three of the slurries, samples weighing ~2 g were weighed to the nearest 0.001 g and evaporated to dryness in an oven at 105°C for 1 hour. High-temperature evaporation was used for the W-A400 and FE-400 slurry components and the W-A400/FE- 400 slurry mixture. After being heated for 1 hour at 105°C, these samples were heated at 500°C for 1 hour to decompose the ferric nitrate to ferric oxide.2 Slurry Trade Abrasive Diluent or Name Component Oxidizer Cab-O-Sperse 1 part SC-1 1.8 parts H2O SC-1 Klebosol 30N50 2 parts 30N50 1 part H2O Semi-Sperse 1 part W-A400 1 part FE-400 W-A400/FE-400 Granite 14 1 part Granite 14A 5.5 parts Granite 14B Table III: Volumetric mix ratios recommended by slurry suppliers. Particle-size distribution measurements were made by the four analyzer vendors listed in Table IV. In all cases, 400-ml samples of dilute slurry were shipped to the instrument manufacturers or their representatives for analysis. The five analyzers used for the study are based on different operating principles and use different algorithms to develop a profile of a sample's particle-size distribution. Some characterize the main distribution or the distribution of particles over a wide range of sizes, while others measure only larger particles. These differences present problems in comparing results from the analyzers. Analyzer Manufacturer Analysis Analysis Model Method Range (µm) CHDF- Matec Applied Capillary 0.015— 2000 Sciences hydrodynamic 1.1 fractionation LA-910 Horiba Classical light 0.02— scattering 1000 Nicomp Particle Sizing Dynamic light 0.003—5 370 Systems scattering AccuSizer Particle Sizing Single-particle 0.5—400 770 Systems light extinction and light scattering LiQuilaz- Particle Single-particle 0.5—20 SO5 Measuring light scattering Systems Table IV: Particle-size distribution analyzers used in the study. Another difference between the particle-size analyzers is the channel width, or range of sizes in each particle count. Instruments with several narrow sizing channels will have fewer particles in each channel than those with fewer and wider channels. Thus, when distribution profiles show the percentage of particles in each channel, the analyzers with narrow channels will have lower peaks than analyzers with wide channels. Channel width is somewhat adjustable on each of the analyzers used for testing. Slurry Percent Density at pH at Component Nonvolatile 20°C (g/ml) 20°C Solids SC-1 30.51 1.194 10.26 30N50 30.14 1.185 10.92 FE-400 W- 2.03 5.97 1.045 1.039 1.19 A400 4.39 Granite 14A 25.78 6.56 1.221 1.037 4.18 Granite 14B 3.92 Table V: Physical properties of slurry components as received from suppliers. Density and pH Results Table V shows the measured density and pH of the individual slurry components as received from the manufacturers; Table VI shows the same properties after preparation of the dilute slurries. In order for density and pH to be used to monitor the concentration of a slurry's nonvolatile abrasive solids, the change in these parameters with a change in percent abrasive concentration must be large enough to be detected reliably by the analytical instruments being used. A linear variation with abrasive concentration was assumed for the dilution range examined. Figure 1 demonstrates the sensitivity of density to abrasive concentration by comparing all four slurries, showing a normalized change in delta density as a function of change in percent abrasive. (Delta density is defined as the deviation in density from the standard mix ratio density.) As the figure shows, the four test slurries behaved very differently. The W-A400/ FE-400 slurry showed almost no change in density as abrasive concentration changed, whereas the 30N50 (dilute) density was very sensitive to changes in abrasive concentration. The other two slurries had intermediate sensitivities. Figure 1: Normalized change in density as a function of percent abrasive for all four test slurries. The relationship of delta density to change in percent abrasive can be explained by the differences between (or similarity of) the densities of the slurry components (see Table V). Because the W-A400 and FE-400 slurry components have nearly identical densities, 1.045 and 1.039 g/ml, respectively, the density of the mixutre does not vary significantly with mix ratio. The densities of the 30N50 slurry concentrate, 1.185 g/ml, and its diluent (water), 1.00 g/ml, are very different; therefore the density of the dilute slurry is very sensitive to the mix ratio. In general, because they are diluted with water, oxide slurries show greater density sensitivity to concentration changes than do tungsten slurries, which are mixtures of an abrasive component and an oxidizer with similar densities. Slurry Percent Density pH Zeta Nonvolatile at at Potential Solids 20°C 20°C (mV) (g/ml) ESA- ZLS- 8000 370 SC-1 12.24 1.066 10.29 —28 —31 (dilute) 30N50 21.10 1.121 10.80 —32 —31 (dilute) W- 4.09 1.042 1.87 5 7 A400/FE- 400 Granite 14 9.80 1.066 4.06 0.3 16 Table VI: Physical properties of slurries after mixing. Although monitoring the density of slurries such as W-A400/FE-400 is not a useful way to monitor mix ratio, for other slurries the change in density might be sufficient to do so. The ability of density to indicate changes in abrasive concentration depends on both the slurry and the precision of the analytical instrument used. Factors that can reduce the reliability of the measurements include the presence of bubbles in the slurry; instrument calibration, repeatability, and drift; and slurry temperature. All of these must be taken into account when determining the useful precision of the instrument. The relationship between instrument resolution, slurry type, and minimum detectable change in abrasive concentration is graphed in Figure 2. As indicated in the figure, a densitometer with a precision of 0.001 g/ml would not effectively detect small changes in the W-A400/FE-400 slurry; the smallest detectable change in percent alumina would be ±0.8%, or ±27% relative error for a nominal 3% abrasive slurry. The densities of the other slurries are more sensitive to changes in abrasive concentration. Using the same densitometer to monitor a Granite 14 slurry with 4.6% abrasives, the resolution of the abrasive concentration would be 4.6% ± 0.2%, or ±4% relative error. Use of a densitometer with higher precision would improve the ability to detect abrasive concentration changes. Figure 2: The effect of densitomter resolution on the minimum detectable change in abrasive concentration. Figure 3: Normalized change in pH as a function of percent abrasive for all four test slurries. Figure 3 shows the change in pH (delta pH) as a function of change in abrasive concentration for all four slurries tested. Comparing the two oxide polishing slurries, the 30N50 (dilute) has a greater pH sensitivity to concentration change than the SC-1 (dilute), which indicates that using pH monitoring as a measure of mix ratio would be more effective for 30N50 than for SC- 1. For the tungsten polishing slurries, the pH change with abrasive concentration for Granite 14 is small because the slurry's two components have a very similar pH, 4.18 and 3.92. In contrast, FE-400 and W-A400 differ by 4 pH units, making the pH sensitivity of that mixture greater. Figure 4: The effect of pH meter resolution on the minimum detectable change in abrasive concentration. Regardless of the pH sensitivities of the slurries, the usefulness of pH monitoring is highly dependent on the precision and reliability of the pH probe and meter. Furthermore, pH stability over a wide range of abrasive concentrations is as important as pH sensitivity to changes in abrasive concentration. The effect of meter resolution on the ability to detect abrasive concentration changes using slurry pH is shown in Figure 4. The SC-1 slurry, the least pH sensitive of those tested, is typically mixed to a nominal 12% silica. If a pH meter with a precision of ±0.01 pH units were used to monitor the dilute slurry, the smallest detectable change in percent silica would be 12% ± 1%, or ±8% relative error. The W-A400/FE-400 slurry is the most pH sensitive; for a slurry mixture with 3% alumina, pH meter resolution would be 3% ± 0.1%, or ±3% relative error. Zeta Potential Results Zeta potential is a measure of the surface charge on a slurry particle. In theory, the higher the absolute zeta potential, the better the slurry stays in suspension. The results of zeta potential measurements by both PSS's ZLS-370 and Matec's ESA-8000 instruments (shown in Table VI) support this theory since the oxide slurries, which are better suspended than the tungsten slurries, exhibited higher absolute zeta potentials (28—32 mV compared with 0.3—16 mV). Particle-Size Distribution Results Figures 5 through 7 show the respective particle-size distribution profiles of the four test slurries from the three analyzers that characterize the main distribution. Because the profiles are displayed in the formats preferred by the analyzer manufacturers, the figures differ in their presentation of the particle data and in the scale of the x-axis (log versus linear). Figure 5: Particle-size distribution analyses of all four test slurries using the CHDF-2000 analyzer. Figure 6: Particle-size distribution analyses of all four test slurries using the LA-910 analyzer. Figure 7: Particle-size distribution analyses of three test slurries using the Nicomp 370 analyzer. (Granite 14 could not be analyzed successfully.) The data in Figure 5 from the CHDF-2000 analyzer are mass weighted. In a mass weighting of a particle-size distribution the masses of all particles in the sample are summed and the percentage of the total mass in each channel calculated. The channels are spaced evenly on a non-log scale, and the data are normalized so that the largest channel percentage is always 100%. However, this does not mean that the number of particles in each slurry's most-populated channel is the same. In Figure 6, the LA-910 analyzer distribution profiles are displayed as the volume-weighted frequency plotted against particle diameter. Similar to mass weighting, volume weighting is accomplished by adding the particle volumes and normalizing the data for the volume in each channel. The channels are spaced equally on a log scale. The data shown in the figure were obtained with the analyzer operating in its standard mode. The LA-910, like some of the other instruments mentioned, can operate in various modes, which allows the detection of smaller peaks, such as particle agglomerates. This multimode operation is accomplished by collecting more data to fit a particular distribution model or by changing model assumptions. The data from the Nicomp 370 analyzer are displayed in Figure 7 as a plot of particle diameter versus volume percent. No results are given for the Granite 14 slurry because the attempted analysis was unsuccessful. Since slurries contain large amounts of solids, they must be diluted in order for most instruments to develop a particle-size distribution. The diluent used should be pH-adjusted to preserve the slurry's stability. Analyzer Slurry CHDF-2000 LA-910 Nicomp 370 Meana Meanb Medianb Meana Mediana Meana Meidana SC-1 128 186 183 175 167 158 135 (dilute) 30N50 74 91 89 85 86 74 67 (dilute) W- 129 204 211 120 110 431 316 A400/FE- 400